DIY Deep Learning for Vision- a Hands-On Tutorial with Caffe.ppt

Feature Visualization Fully Convolutional Models Transform fixed-input models into any-size models by translating inner products to convolutions. The computation exploits a natural efficiency of convolutional neural network (CNN) structure by dynamic programming in the forward pass from shallow to deep layers and analogously in backward. Net surgery in Caffe how to transform models: - make fully convolutional - transplant parameters [ OverFeat] RELATED PROJECTS Object Detection R-CNN: Regions with Convolutional Neural Networks /github/BVLC/caffe/blob/master/examples/detection.ipynb Full R-CNN scripts available at /rbgirshick/rcnn Ross Girshick et al. Rich feature hierarchies for accurate object detection and semantic segmentation. CVPR14. Visual Style Recognition Other Styles: Vintage Long Exposure Noir Pastel Macro … and so on. Karayev et al. Recognizing Image Style. BMVC14. Caffe fine-tuning example. Demo online at / (see Results Explorer). [ Image-Style] Documentation - all tutorial materials - long-form descriptions - developer API LATEST ROAST Caffe + cuDNN - NVIDIA-Caffe collaboration - drop-in acceleration library for key computations - convolution, pooling, nonlinearities, softmax - 1.2 - 3x layer-wise speedups - in release candidate edition of master branch now LATEST ROAST ** not including IO Parallel Caffe - parallel / distributed training across CPUs, GPUs,and cluster nodes - pull request from Flickr by @cypof (thanks!)- promoted to official branch for integration in PR #1148 - join the work-in-progress to help merge LATEST ROAST LAST SIP Caffe... is fast is state-of-the-art has tips, recipes, demos, and models brings together an active community ...all for free and open source Yangqing Jia, Evan Shelhamer, Jeff Donahue, Sergey Karayev Jonathan Long, Ross Girshick, Sergio Guadarrama Thanks to the Caffe crew ...plus the cold-brew and our open source contributors! Acknowledgements Thank you to the Berkeley Vision and Learning Center Sponsors. T